Adversarial Attacks are Reversible with Natural Supervision | IEEE Conference Publication | IEEE Xplore